pySCENIC protocol: PBMC10k downstream analyses¶
August 2019
Dataset: 10k PBMCs from a Healthy Donor available from 10x Genomics (here).
This notebook uses a loom file generated from the first part of the SCENIC protocol, described in: PBMC10k_SCENIC-protocol-CLI.ipynb
In [1]:
# import dependencies
import os
import numpy as np
import pandas as pd
import scanpy as sc
import loompy as lp
from MulticoreTSNE import MulticoreTSNE as TSNE
import json
import base64
import zlib
from pyscenic.plotting import plot_binarization
from pyscenic.export import add_scenic_metadata
from pyscenic.cli.utils import load_signatures
import matplotlib as mpl
import matplotlib.pyplot as plt
from scanpy.plotting._tools.scatterplots import plot_scatter
import seaborn as sns
set variables for file paths to read from and write to:¶
In [2]:
# set a working directory
wdir = "/mnt/beegfs/data/users/vsc32528/pbmc10kprotocol/"
wdir = '/ddn1/vol1/staging/leuven/stg_00002/lcb/cflerin/testruns/scenic-protocol/prottest4'
os.chdir( wdir )
# path to loom output, generated from a combination of Scanpy and pySCENIC results:
f_final_loom = 'PBMC10k.loom'
In [3]:
sc.settings.verbosity = 3 # verbosity: errors (0), warnings (1), info (2), hints (3)
sc.logging.print_versions()
sc.settings.set_figure_params(dpi=150)
Extract relevant data from the integrated loom file¶
In [4]:
# scenic output
lf = lp.connect( f_final_loom, mode='r', validate=False )
meta = json.loads(zlib.decompress(base64.b64decode( lf.attrs.MetaData )))
exprMat = pd.DataFrame( lf[:,:], index=lf.ra.Gene, columns=lf.ca.CellID).T
auc_mtx = pd.DataFrame( lf.ca.RegulonsAUC, index=lf.ca.CellID)
In [5]:
# create a dictionary of regulons:
regulons = {}
for i,r in pd.DataFrame(lf.ra.Regulons,index=lf.ra.Gene).iteritems():
regulons[i] = list(r[r==1].index.values)
In [6]:
# cell annotations from the loom column attributes:
cellAnnot = pd.concat(
[
pd.DataFrame( lf.ca.Celltype_Garnett, index=lf.ca.CellID ),
pd.DataFrame( lf.ca.ClusterID, index=lf.ca.CellID ),
pd.DataFrame( lf.ca.Louvain_clusters_Scanpy, index=lf.ca.CellID ),
pd.DataFrame( lf.ca.Percent_mito, index=lf.ca.CellID ),
pd.DataFrame( lf.ca.nGene, index=lf.ca.CellID ),
pd.DataFrame( lf.ca.nUMI, index=lf.ca.CellID ),
],
axis=1
)
cellAnnot.columns = [
'Celltype_Garnett',
'ClusterID',
'Louvain_clusters_Scanpy',
'Percent_mito',
'nGene',
'nUMI']
In [7]:
# capture embeddings:
dr = [
pd.DataFrame( lf.ca.Embedding, index=lf.ca.CellID )
]
dr_names = [
meta['embeddings'][0]['name'].replace(" ","_")
]
# add other embeddings
drx = pd.DataFrame( lf.ca.Embeddings_X, index=lf.ca.CellID )
dry = pd.DataFrame( lf.ca.Embeddings_Y, index=lf.ca.CellID )
for i in range( len(drx.columns) ):
dr.append( pd.concat( [ drx.iloc[:,i], dry.iloc[:,i] ], sort=False, axis=1, join='outer' ))
dr_names.append( meta['embeddings'][i+1]['name'].replace(" ","_").replace('/','-') )
# rename columns:
for i,x in enumerate( dr ):
x.columns = ['X','Y']
In [8]:
lf.close()
Alternately, we can load this data into a scanpy.AnnData object¶
This can be done directly from the integrated loom file, with a few modifications to allow for SCENIC- and SCope-specific loom attributes:
In [9]:
adata = sc.read( f_final_loom, validate=False)
# drop the embeddings and extra attributes from the obs object
adata.obs.drop( ['Embedding','Embeddings_X','Embeddings_Y','RegulonsAUC'], axis=1, inplace=True )
In [10]:
# add the embeddings into the adata.obsm object
for i,x in enumerate( dr ):
adata.obsm[ 'X_'+dr_names[i] ] = x.as_matrix()
In [11]:
sc.utils.sanitize_anndata( adata )
We can also add all metadata derived from SCENIC to the scanpy.AnnData object.¶
In [ ]:
# load the regulons from a file using the load_signatures function
sig = load_signatures('reg.csv')
add_scenic_metadata(adata, auc_mtx, sig)
Display a motifs table with motif logos¶
View the motifs table along with motif logos¶
In [12]:
# helper functions (not yet integrated into pySCENIC):
from pyscenic.utils import load_motifs
import operator as op
from IPython.display import HTML, display
BASE_URL = "http://motifcollections.aertslab.org/v9/logos/"
COLUMN_NAME_LOGO = "MotifLogo"
COLUMN_NAME_MOTIF_ID = "MotifID"
COLUMN_NAME_TARGETS = "TargetGenes"
def display_logos(df: pd.DataFrame, top_target_genes: int = 3, base_url: str = BASE_URL):
"""
:param df:
:param base_url:
"""
# Make sure the original dataframe is not altered.
df = df.copy()
# Add column with URLs to sequence logo.
def create_url(motif_id):
return '<img src="{}{}.png" style="max-height:124px;"></img>'.format(base_url, motif_id)
df[("Enrichment", COLUMN_NAME_LOGO)] = list(map(create_url, df.index.get_level_values(COLUMN_NAME_MOTIF_ID)))
# Truncate TargetGenes.
def truncate(col_val):
return sorted(col_val, key=op.itemgetter(1))[:top_target_genes]
df[("Enrichment", COLUMN_NAME_TARGETS)] = list(map(truncate, df[("Enrichment", COLUMN_NAME_TARGETS)]))
MAX_COL_WIDTH = pd.get_option('display.max_colwidth')
pd.set_option('display.max_colwidth', 200)
display(HTML(df.head().to_html(escape=False)))
pd.set_option('display.max_colwidth', MAX_COL_WIDTH)
In [13]:
df_motifs = load_motifs('reg.csv')
In [14]:
selected_motifs = ['PAX5','TCF3','EBF1']
df_motifs_sel = df_motifs.iloc[ [ True if x in selected_motifs else False for x in df_motifs.index.get_level_values('TF') ] ,:]
In [15]:
#display_logos(df_motifs.head())
display_logos( df_motifs_sel.sort_values([('Enrichment','NES')], ascending=False).head(9))
| Enrichment | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| AUC | Annotation | Context | MotifSimilarityQvalue | NES | OrthologousIdentity | RankAtMax | TargetGenes | MotifLogo | ||
| TF | MotifID | |||||||||
| EBF1 | transfac_pro__M07351 | 0.153960 | gene is directly annotated | (hg38__refseq-r80__500bp_up_and_100bp_down_tss.mc9nr, top50, activating) | 0.000000 | 6.271401 | 1.0 | 612 | [(FAM129C, 10.255336830422989), (CD79A, 10.822303190915058), (PAX5, 12.393443190322412)] |  |
| factorbook__EBF1 | 0.138138 | gene is directly annotated | (hg38__refseq-r80__500bp_up_and_100bp_down_tss.mc9nr, top50, activating) | 0.000000 | 5.476125 | 1.0 | 756 | [(CD79A, 10.255336830422989), (FAM129C, 10.822303190915058), (P2RY10, 12.393443190322412)] |  |
|
| PAX5 | transfac_pro__M04711 | 0.142637 | gene is annotated for similar motif dbcorrdb__PAX5__ENCSR000BHJ_1__m2 ('PAX5 (ENCSR000BHJ-1, motif 2)'; q-value = 0.000317) | (hg38__refseq-r80__500bp_up_and_100bp_down_tss.mc9nr, top50, activating) | 0.000317 | 5.273931 | 1.0 | 292 | [(CNR2, 25.280202022610478), (KIAA0040, 27.735438642040428), (STAP1, 32.35398721525476)] |  |
| cisbp__M1907 | 0.141739 | gene is annotated for similar motif dbcorrdb__PAX5__ENCSR000BHJ_1__m2 ('PAX5 (ENCSR000BHJ-1, motif 2)'; q-value = 0.000547) | (hg38__refseq-r80__500bp_up_and_100bp_down_tss.mc9nr, top50, activating) | 0.000547 | 5.230762 | 1.0 | 2490 | [(CNR2, 23.63296995309573), (E2F5, 24.70069302305977), (CD37, 25.280202022610478)] |  |
|
| dbcorrdb__SPI1__ENCSR000BGW_1__m1 | 0.140306 | gene is annotated for similar motif dbcorrdb__PAX5__ENCSR000BHJ_1__m2 ('PAX5 (ENCSR000BHJ-1, motif 2)'; q-value = 0.000405) | (hg38__refseq-r80__500bp_up_and_100bp_down_tss.mc9nr, top50, activating) | 0.000405 | 5.161844 | 1.0 | 1596 | [(ARHGAP24, 24.70069302305977), (CD22, 25.280202022610478), (BIRC3, 27.735438642040428)] |  |
|
Dimensionality reduction plots¶
Show both highly variable genes and SCENIC UMAP plots with Louvain clustering and cell type classifications¶
In [16]:
sc.set_figure_params(frameon=False, dpi=150, fontsize=10)
sc.pl.scatter( adata, basis='highly_variable_genes_UMAP',
color=['Louvain_clusters_Scanpy','Celltype_Garnett'],
title=['HVG - UMAP (Louvain clusters)','HVG - UMAP (Cell type)'],
alpha=0.8,
save='_Louvain-celltype.svg'
)
sc.pl.scatter( adata, basis='SCENIC_AUC_UMAP',
color=['Louvain_clusters_Scanpy','Celltype_Garnett'],
title=['SCENIC - UMAP (Louvain clusters)','SCENIC - UMAP (Cell type)'],
alpha=0.8,
save='_Louvain-celltype.svg'
)
Alternately, we can plot two dimensionality reductions side-by-side¶
(this uses non-Scanpy plotting functions)
In [17]:
def colorMap( x, palette='bright' ):
import natsort
from collections import OrderedDict
#
n=len(set(x))
cpalette = sns.color_palette(palette,n_colors=n )
cdict = dict( zip( list(set(x)), cpalette ))
cmap = [ cdict[i] for i in x ]
cdict = OrderedDict( natsort.natsorted(cdict.items()) )
return cmap,cdict
def drplot( dr, colorlab, ax, palette='bright', title=None, **kwargs ):
cmap,cdict = colorMap( colorlab, palette )
for lab,col in cdict.items():
ix = colorlab.loc[colorlab==lab].index
ax.scatter( dr['X'][ix], dr['Y'][ix], c=[col]*len(ix), alpha=0.7, label=lab, edgecolors='none')
if( title is not None ):
ax.set_title(title, fontsize='x-large');
#
ax.set_xticks([])
ax.set_yticks([])
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
ax.spines['bottom'].set_visible(False)
ax.spines['left'].set_visible(False)
In [18]:
plt.rcParams.update({'font.size':12})
fig, (ax1,ax2) = plt.subplots(1,2, figsize=(20,10), dpi=150 )
drplot( dr[0], colorlab=cellAnnot['Louvain_clusters_Scanpy'], ax=ax1, palette='bright', s=2, title='Highly variable genes - UMAP' )
drplot( dr[4], colorlab=cellAnnot['Louvain_clusters_Scanpy'], ax=ax2, palette='bright', s=2, title='SCENIC AUC - UMAP' )
ax2.legend(loc='right', bbox_to_anchor=(1.15, 0.5), ncol=1, markerscale=2, fontsize='x-large', frameon=False, title="Louvain\nclusters")
plt.tight_layout()
plt.savefig("PBMC10k_dimred_umap-hvg-scenic-louvain.png", dpi=150, bbox_inches = "tight")
Regulon specificity scores (RSS) across predicted cell types¶
In [19]:
from pyscenic.rss import regulon_specificity_scores
from pyscenic.plotting import plot_rss
import matplotlib.pyplot as plt
from adjustText import adjust_text
import seaborn as sns
from pyscenic.binarization import binarize
Calculate RSS¶
In [20]:
rss_cellType = regulon_specificity_scores( auc_mtx, cellAnnot['Celltype_Garnett'] )
rss_cellType
Out[20]:
RSS panel plot with all cell types¶
In [21]:
cats = sorted(list(set(cellAnnot['Celltype_Garnett'])))
fig = plt.figure(figsize=(15, 8))
for c,num in zip(cats, range(1,len(cats)+1)):
x=rss_cellType.T[c]
ax = fig.add_subplot(2,5,num)
plot_rss(rss_cellType, c, top_n=5, max_n=None, ax=ax)
ax.set_ylim( x.min()-(x.max()-x.min())*0.05 , x.max()+(x.max()-x.min())*0.05 )
for t in ax.texts:
t.set_fontsize(12)
ax.set_ylabel('')
ax.set_xlabel('')
adjust_text(ax.texts, autoalign='xy', ha='right', arrowprops=dict(arrowstyle='-',color='lightgrey'), precision=0.001 )
fig.text(0.5, 0.0, 'Regulon', ha='center', va='center', size='x-large')
fig.text(0.00, 0.5, 'Regulon specificity score (RSS)', ha='center', va='center', rotation='vertical', size='x-large')
plt.tight_layout()
plt.rcParams.update({
'figure.autolayout': True,
'figure.titlesize': 'large' ,
#'axes.labelsize': 'x-large',
#'axes.titlesize':'x-large',
'xtick.labelsize':'large',
'ytick.labelsize':'large'
})
plt.savefig("PBMC10k_cellType-RSS-top5.png", dpi=150, bbox_inches = "tight")
plt.show()
Select the top 5 regulons from each cell type¶
In [22]:
topreg = []
for i,c in enumerate(cats):
topreg.extend(
list(rss_cellType.T[c].sort_values(ascending=False)[:5].index)
)
topreg = list(set(topreg))
Generate a Z-score for each regulon to enable comparison between regulons¶
In [23]:
auc_mtx_Z = pd.DataFrame( index=auc_mtx.index )
for col in list(auc_mtx.columns):
auc_mtx_Z[ col ] = ( auc_mtx[col] - auc_mtx[col].mean()) / auc_mtx[col].std(ddof=0)
#auc_mtx_Z.sort_index(inplace=True)
Generate a heatmap¶
In [24]:
def palplot(pal, names, colors=None, size=1):
n = len(pal)
f, ax = plt.subplots(1, 1, figsize=(n * size, size))
ax.imshow(np.arange(n).reshape(1, n),
cmap=mpl.colors.ListedColormap(list(pal)),
interpolation="nearest", aspect="auto")
ax.set_xticks(np.arange(n) - .5)
ax.set_yticks([-.5, .5])
ax.set_xticklabels([])
ax.set_yticklabels([])
colors = n * ['k'] if colors is None else colors
for idx, (name, color) in enumerate(zip(names, colors)):
ax.text(0.0+idx, 0.0, name, color=color, horizontalalignment='center', verticalalignment='center')
return f
In [25]:
colors = sns.color_palette('bright',n_colors=len(cats) )
colorsd = dict( zip( cats, colors ))
colormap = [ colorsd[x] for x in cellAnnot['Celltype_Garnett'] ]
In [26]:
sns.set()
sns.set(font_scale=0.8)
fig = palplot( colors, cats, size=1.0)
In [27]:
g = sns.clustermap(auc_mtx_Z[topreg], annot=False, square=False, linecolor='gray',
yticklabels=False, vmin=-2, vmax=6, row_colors=colormap,
cmap="YlGnBu", figsize=(21,16) )
sns.set(font_scale=0.9)
g.cax.set_visible(True)
g.ax_heatmap.set_ylabel('')
g.ax_heatmap.set_xlabel('')
Out[27]:
Generate a binary regulon activity matrix:¶
In [28]:
binary_mtx, auc_thresholds = binarize( auc_mtx, num_workers=25 )
binary_mtx.head()
Out[28]:
Show the AUC distributions for selected regulons¶
In [29]:
# select regulons:
r = [ 'RXRA_(+)', 'NFE2_(+)', 'ETV2_(+)' ]
fig, axs = plt.subplots(1, 3, figsize=(16, 5), dpi=150, sharey=False)
for i,ax in enumerate(axs):
sns.distplot(auc_mtx[ r[i] ], ax=ax, norm_hist=True, bins=100)
ax.plot( [ auc_thresholds[ r[i] ] ]*2, ax.get_ylim(), 'r:')
ax.title.set_text( r[i] )
ax.set_xlabel('')
fig.text(0.00, 0.5, 'Frequency', ha='center', va='center', rotation='vertical', size='x-large')
fig.text(0.5, 0.0, 'AUC', ha='center', va='center', rotation='horizontal', size='x-large')
fig.tight_layout()
fig.savefig("PBMC10k_cellType-binaryPlot2.png", dpi=150, bbox_inches = "tight")
Regulon specificity scores (RSS) across Louvain clusters¶
Calculate RSS¶
In [30]:
rss_louvain = regulon_specificity_scores( auc_mtx, cellAnnot['Louvain_clusters_Scanpy'] )
rss_louvain
Out[30]:
RSS panel plot with all cell types¶
In [32]:
cats = sorted( list(set(cellAnnot['Louvain_clusters_Scanpy'])), key=int )
fig = plt.figure(figsize=(15, 12))
for c,num in zip(cats, range(1,len(cats)+1)):
x=rss_louvain.T[c]
ax = fig.add_subplot(3,5,num)
plot_rss(rss_louvain, c, top_n=5, max_n=None, ax=ax)
ax.set_ylim( x.min()-(x.max()-x.min())*0.05 , x.max()+(x.max()-x.min())*0.05 )
for t in ax.texts:
t.set_fontsize(12)
ax.set_ylabel('')
ax.set_xlabel('')
adjust_text(ax.texts, autoalign='xy', ha='right', arrowprops=dict(arrowstyle='-',color='lightgrey'), precision=0.001 )
fig.text(0.5, 0.0, 'Regulon', ha='center', va='center', size='x-large')
fig.text(0.00, 0.5, 'Regulon specificity score (RSS)', ha='center', va='center', rotation='vertical', size='x-large')
plt.tight_layout()
plt.rcParams.update({
'figure.autolayout': True,
'figure.titlesize': 'large' ,
'xtick.labelsize':'large',
'ytick.labelsize':'large'
})
plt.savefig("PBMC10k_Louvain-RSS-top5.png", dpi=150, bbox_inches = "tight")
plt.show()
Select the top 5 regulons from each cell type¶
In [33]:
topreg = []
for i,c in enumerate(cats):
topreg.extend(
list(rss_louvain.T[c].sort_values(ascending=False)[:5].index)
)
topreg = list(set(topreg))
Generate a heatmap¶
In [34]:
def palplot(pal, names, colors=None, size=1):
n = len(pal)
f, ax = plt.subplots(1, 1, figsize=(n * size, size))
ax.imshow(np.arange(n).reshape(1, n),
cmap=mpl.colors.ListedColormap(list(pal)),
interpolation="nearest", aspect="auto")
ax.set_xticks(np.arange(n) - .5)
ax.set_yticks([-.5, .5])
ax.set_xticklabels([])
ax.set_yticklabels([])
colors = n * ['k'] if colors is None else colors
for idx, (name, color) in enumerate(zip(names, colors)):
ax.text(0.0+idx, 0.0, name, color=color, horizontalalignment='center', verticalalignment='center')
return f
In [36]:
colors = sns.color_palette('bright',n_colors=len(cats) )
colorsd = dict( zip( cats, colors ))
colormap = [ colorsd[x] for x in cellAnnot['Louvain_clusters_Scanpy'] ]
In [37]:
sns.set()
sns.set(font_scale=0.8)
fig = palplot( colors, cats, size=1.0)
In [38]:
g = sns.clustermap(auc_mtx_Z[topreg], annot=False, square=False, linecolor='gray',
yticklabels=False, vmin=-2, vmax=6, row_colors=colormap,
cmap="YlGnBu", figsize=(21,16) )
sns.set(font_scale=0.9)
g.cax.set_visible(True)
g.ax_heatmap.set_ylabel('')
g.ax_heatmap.set_xlabel('')
Out[38]:
Further exploration of modules directly from the network inference output¶
In [39]:
adjacencies = pd.read_csv("adj.tsv", index_col=False, sep='\t')
Create the modules:
In [40]:
from pyscenic.utils import modules_from_adjacencies
modules = list(modules_from_adjacencies(adjacencies, exprMat))
pick out modules for EBF1:¶
In [41]:
tf = 'EBF1'
tf_mods = [ x for x in modules if x.transcription_factor==tf ]
for i,mod in enumerate( tf_mods ):
print( f'{tf} module {str(i)}: {len(mod.genes)} genes' )
print( f'{tf} regulon: {len(regulons[tf+"_(+)"])} genes' )
write these modules, and the regulon to files:
In [42]:
for i,mod in enumerate( tf_mods ):
with open( tf+'_module_'+str(i)+'.txt', 'w') as f:
for item in mod.genes:
f.write("%s\n" % item)
with open( tf+'_regulon.txt', 'w') as f:
for item in regulons[tf+'_(+)']:
f.write("%s\n" % item)
In [43]:
from IPython.display import display, Image
display(Image(filename='iRegulon_screenshot_PBMC10k-EBF1.png'))
